1

git@github.com:PlayVoice/vits_chinese.git

1
2
3
4
5

pip install -r requirements.txt
cd monotonic_align
python setup.py build_ext --inplace

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82

import mtutils as mt
import pypinyin
def make_chinese_label(xf_res):
    chinese_label = ''
    pure_chinese_str_list = list()
    chinese_char_num = 0
    for sub_sentence in xf_res[:-1]:
        pure_chinese_str = ''
        for sub_data in sub_sentence:
            cw = sub_data['cw']
            assert len(cw) == 1
            cw = cw[0]
            if len(cw['w']) > 1:
                chinese_label += cw['w']
                chinese_label += '#2'
                pure_chinese_str += cw['w']
            else:
                single_char = cw['w']
                if single_char in [',', '，', '、', '：', '?','？', '！', '!']:
                    chinese_label = chinese_label[:-1] + '3'
                    chinese_label += single_char
                else:
                    chinese_label += single_char
                    chinese_label += '#1'
                    pure_chinese_str += single_char
        pure_chinese_str_list.append(pure_chinese_str)
        chinese_char_num += len(pure_chinese_str)
    chinese_label = chinese_label[:-1] + '4'
    chinese_label += xf_res[-1][0]['cw'][0]['w']
    return chinese_label, pure_chinese_str_list, chinese_char_num
def make_pinyin_label(pure_chinese_str_list):
    sub_res_str_list = list()
    for pure_chinese_str in pure_chinese_str_list:
        pinyin_sub_res = pypinyin.lazy_pinyin(pure_chinese_str, style=pypinyin.STYLE_TONE3, neutral_tone_with_five=True)
        sub_res_str = mt.str_connection(*pinyin_sub_res, connect_char=' ')
        sub_res_str_list.append(sub_res_str)
    result_str = mt.str_connection(*sub_res_str_list, connect_char=' ')
    return result_str
if __name__ == '__main__':
    role_name = '云堇'
    dir_list, file_list = mt.get_dir_file_list('fea/classified_data_plus/')
    match_num = 0
    match_dir_path = None
    for dir_path in dir_list:
        if role_name in dir_path:
            match_num += 1
            match_dir_path = dir_path
    assert match_num == 1
    print(match_dir_path)
    xf_match_res = mt.json_load('xf_rec_results.json')
    role_wav_path_list = mt.glob_recursively(match_dir_path, 'wav')
    match_pair = list()
    loss_pair = list()
    for role_wav_path in mt.tqdm(role_wav_path_list):
        key = mt.OS_basename(role_wav_path)
        if key in xf_match_res:
            match_pair.append([key, xf_match_res[key]])
        else:
            loss_pair.append(key)
        label_result_str_list = list()
        for match_item in mt.tqdm(match_pair, desc="生成标注： "):
        key = mt.get_path_stem(match_item[0]).replace(' ', '_')
        data = match_item[1]['data']
        chinese_label, pure_chinese_str_list, chinese_char_num = make_chinese_label(data)
        first_line = key + '\t' + chinese_label
        second_line = '\t' + make_pinyin_label(pure_chinese_str_list)
        if chinese_char_num > 2:
            label_result_str_list.append(first_line)
            label_result_str_list.append(second_line)
    mt.file_write_lines(label_result_str_list, 'gt_label/' + role_name + '_gt_1.0.txt', overwrite=True)
    pass

1
2
3
4
5
6

3011_00035	不然#2可#1赶#1不#1完#1每日#2的#1功课#4。
	bu4 ran2 ke3 gan3 bu4 wan2 mei3 ri4 de5 gong1 ke4
1013_00252	小时候#3，我#1随#1父亲#2来过#2一次#2天#1恒#1山#3，记得#2这#1座#1山#1好#1高#1好#1高#3，爬到#2山顶#2时#1腿#1已经#2酸#1的#1不行#2了#4。
	xiao3 shi2 hou4 wo3 sui2 fu4 qin1 lai2 guo4 yi1 ci4 tian1 heng2 shan1 ji4 de5 zhe4 zuo4 shan1 hao3 gao1 hao3 gao1 pa2 dao4 shan1 ding3 shi2 tui3 yi3 jing1 suan1 de5 bu4 xing2 le5
3011_00085	啊#3，你们#2是#1哪#1家#1报社#2的#3？为什么#2之前#2都#1没#1见过#4？
	a5 ni3 men5 shi4 na3 jia1 bao4 she4 de5 wei4 shen2 me5 zhi1 qian2 dou1 mei2 jian4 guo4

1

python vits_prepare.py -c ./configs/bert_vits.json

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

def main():
    """Assume Single Node Multi GPUs Training Only"""
    assert torch.cuda.is_available(), "CPU training is not allowed."
    n_gpus = torch.cuda.device_count()
    os.environ["MASTER_ADDR"] = "localhost"
    os.environ["MASTER_PORT"] = "40000"
    hps = utils.get_hparams()
    run(0, 1, hps)
    # mp.spawn(
    #     run,
    #     nprocs=n_gpus,
    #     args=(
    #         n_gpus,
    #         hps,
    #     ),
    # )

1

python train.py -c configs/bert_vits.json -m bert_vits

1
2

download baker data: https://www.data-baker.com/data/index/TNtts/
change sample rate of waves to 16kHz, and put waves to ./data/waves

1
2
3

遥望星空作文独自坐在乡间的小丘上，看着阳光渐渐变暗，听着鸟鸣渐渐变弱，触着清风渐渐变凉。
时光总是慢慢地偷走我们的容颜，渐渐地有些人终将离我们而远去。
白色的樱花纯洁高尚，红色的樱花热烈奔放，绿色的樱花清晰澹雅，花开的美丽与快乐，花落的烂漫与潇洒都蕴藏着樱花的人生智慧。

1

Original error was: DLL load failed while importing _multiarray_umath: 找不到指定的模块。

1

ValueError: check_hostname requires server_hostname

1

AttributeError: module ‘distutils‘ has no attribute ‘version‘

1

pip uninstall setuptools

1

pip install setuptools==58.0.4

1

python setup.py build_ext --inplace

1

D:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\bin\Hostx86\x64

1

fatal error C1356: 无法找到 mspdbcore.dll

1

python setup.py build_ext --inplace --include-dirs="F:\Windows Kits\10\Include\10.0.17763.0\ucrt;D:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\include;F:\Windows Kits\10\Include\10.0.17763.0\shared" --library-dirs="D:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\lib\x64;F:\Windows Kits\10\Lib\10.0.17763.0\um\x64;F:\Windows Kits\10\Lib\10.0.17763.0\ucrt\x64;F:\Windows Kits\10\bin\10.0.17763.0\x64"